Real-world visual–inertial and acoustic simultaneous localization and mapping for underwater remotely operated vehicle navigation

Abstract

This work presents real-world experimental results from a Bayesian underwater simultaneous localization and mapping (SLAM) system deployed on a remotely operated vehicle (ROV) and evaluated in both controlled pool experiments and ocean settings. The SLAM system uses a feature-based visual SLAM approach that extracts repeatable keypoints and descriptors from a forward-facing camera and matches them across frames. These tracked visual features are fused with inertial measurements in a Maximum a Posteriori optimization to jointly estimate the ROV trajectory (pose over time) and a sparse visual landmark map. Camera calibration is performed to estimate the intrinsic parameters and lens distortion required for accurate feature projection. Acoustic sensing is integrated through an altimeter and sonar measurements, providing range-to-seafloor and to landmarks, enhancing vertical motion and overall 3-D trajectory estimation in visually degraded underwater conditions. System performance in the ocean setting is evaluated using fixed, moored reference points with known GPS locations to provide ground-truth trajectory validation.